Giacomo Bignardi

as I'm revising my course materials, I keep stumbling upon cool @mc-stan.org developments.

Current favorites:
1. your model has funnels and you exhausted reparametrization ideas: metric = "dense_e" makes your HMC learn about covariance btw parameters. Sloooow, but effective!
1/

Worth noting this approach works with python/jupyter aswell, and across windows/linux/macos. Also RAM prices will be crazy in 2026 due to AI, so prices will change!

But you can also buy a windows/linux desktop or mini PC with lots of RAM for far less money, if you don't mind slightly worse CPU performance. The differences are even bigger if you need 128GB of RAM....

Apple users who need 64GB of RAM will need to shell out either £3,879 for a MacBook 16 Pro or £1,839 for a Mac mini. Therefore, you could get both a 64GB mini and a basic MacBook Air (£900) for much less - and you'll only lose £900 when you leave your laptop on the tube.

If you already have a gaming computer at home, with Docker + Tailscale, you can work from it remotely pretty seamlessly. Alternatively, if you only occasionally need more computer power, you can rent a server (running Docker) from your university or a private company (e.g., www.hetzner.com/cloud).

For illustrative purposes, I've installed tailscale on my desktop and phone. On my phone, I simply replace the localhost part of the web address with the tailscale ip address of my desktop, and bingo, I can access RStudio server on another device - and it works largely the same.

A nice thing about Docker (though this isn't exclusive to Docker) is that it runs RStudio Server, which we access via a web browser. By default, we can only access it on our own computer, but we can also access it from other computers via a secure, encrypted connection using the free app Tailscale.

Most of us work partly from home, so we need to work in different locations. For most, this means buying a laptop. However, if you work with large datasets and do complex analyses, you may need a higher-spec laptop with more RAM, which can cost 2-3x as much as the basic-spec version.

Hot take: one of the main benefits of running data analyses in 🐬 Docker 🐬 isn't reproducibility, but saving yourself (or your lab) hundreds or thousands of pounds in computer costs*. Here's how... #datascience #docker #rstudio #python

*for a very niche set of people

YouTube video by C-SPAN Doctor on How Screen Time Hurts Kids' Cognitive Development

video making very strong (& problematic) claims about the supposed damaging effects of screen time on children's education & cognition is trending on youtube (from the US senate committee)... www.youtube.com/watch?v=Fd-_...

...claims such as: "Gen Z is the first population to cognitively underperform us on every measure", "Once countries widely adopt digital technologies in school, performance goes down wildly.", "We have evolved biologically to learn from other human beings, not from screens."

YouTube video by C-SPAN Doctor on How Screen Time Hurts Kids' Cognitive Development

video making very strong (& problematic) claims about the supposed damaging effects of screen time on children's education & cognition is trending on youtube (from the US senate committee)... www.youtube.com/watch?v=Fd-_...

Michael Betancourt - Chief Research Scientist, Symplectomorphic, LLC Generative ModelingOver the last few years the term "generative modeling" has become increasingly popular.  The precise definition of the term, however, tends to be vague and often inconsistent across...

On Tues, February 17th I'll be giving a talk and short workshop on (narratively) generative modeling for Princeton's psychology department, psychology.princeton.edu/news-events/.... The talk (and maybe the workshop) will be live-streamed for those interested in attending remotely.

I've used Lavaan almost every (work)day for 15 years, but this open source labour of love has never received proper institutional support. I'm delighted to be a small part of an 1.5M OpenScienceNL award, led by Jorgensen, to completely revamp and futureproof Lavaan www.openscience.nl/en/news/45-p...

For those who have recieved non taxable phd stipends and have been in employment directly before or after, you may have overpaid your student loan due to the sneeky way it's collected...

Student loan overpayments This MoneySavingExpert guide can help you check if you've overpaid your student loan and reclaim the cash if so. 10,000s overpay their student loans – have you?

Wow if you have a UK student loan this is definitely worth checking, Xmas is on @martinlewis.moneysavingexpert.com again this year 🎄🎄https://www.moneysavingexpert.com/students/student-loan-overpayment-refund

In 2021 we reported that live learning outperformed recorded learning. In a new preregistered analysis, my first senior-author paper led by Stan de Visser (pre-print), we find that this benefit does not increase with interactivity. The potential to interact may be enough to boost learning. A thread:

Happy to share my new paper published in @nathumbehav.nature.com: A critical look at statistical power in computational modeling studies, particularly those based on model selection.
www.nature.com/articles/s41...

Dont forget the true score isnt the actual "true" score, but the expected score.

Link doesn't work for me sadly! :( Says bad gateway error?

A "methods primer" article in the journal "BMJ Medicine", titled "Factors associated with: problems of using exploratory multivariable regression to identify causal risk factors"

We wrote an article explaining why you shouldn't put several variables into a regression model and report which are statistically significant - even as exploratory research. bmjmedicine.bmj.com/content/4/1/.... How did we do?

Interesting, thanks! I need to read up on this again as I don't remember all the different ICCs. Cool that it seems to match up with this formula!

... which would converge to 1 as the number of raters increases? I assume with inter-rater reliability, researchers are interested in something like how well a single rater's ratings of the items match the "true" rating, which is quite different, and necessitates a different approach?

Hmmm - I haven't really thought about inter-rater reliability since 2016 - so I'm not sure how RMU would map onto existing metrics yet. Thinking aloud: applying RMU to the item intercepts gives the reliability of the model's item estimates when combining all information across all raters...

New paper by brilliant final yer PhD student Blanca Piera Pi-Sunyer and colleagues, including 3 undergraduate students (Evans, Ratcliffe, Janaarthanan) 🤩

The Relationship between Perceived Friendship Quality and Self-Judgements in Adolescent Girls from London

journals.sagepub.com/doi/epub/10....

Blanca Piera Pi-Sunyer on how to make the world a better place for young people, the cumulative impact of adversity & how to mitigate it in the Gates Cambridge podcast, So, now what? - @cambridgeuni.bsky.social

Could you share the models/code your using? Do you have estimates for subjects that you plug into the function?

Neat! I have brms code for replicating (and going beyond) alpha too incase its useful: www.bignardi.co.uk/8_bayes_reli...

Hmm, interesting, I've had this on my reading list but have not read it yet, the intro is so clear and great! I disagree that reliability implies equal precision across scores, but it seems like an interesting way of going beyond reliability...

Good point, would be interesting to try (and also force myself to learn generalizability theory in more detail)!